Building equitable and feasible clinical decision support tools for population health management of people living with chronic lung disease - PROJECT SUMMARY People living with chronic obstructive lung disease, interstitial lung disease, and other forms of chronic lung disease frequently experience exacerbations requiring acute care hospitalizations. These severe exacerba- tions requiring hospitalization are associated with decreased quality of life for patients, increased morbidity and mortality, and increased burdens on health system resources. Through work funded by my K23 award, I have identified actionable risk mechanisms that contribute to hospitalization risk in this population. Notably, many of these mechanisms are present for months or even years among community-dwelling patients liv- ing with chronic lung disease prior to a hospitalization. These findings suggest an opportunity to identify patients with these actionable risks and refer them for appropriate interventions as early as possible. To accomplish this, my K23 supported the development of two prediction models that can efficiently analyze electronic health record data, including both structured (e.g. demographics, diagnoses, and laboratory tests) and unstructured sources (e.g. the text of clinical encounter notes), of large populations to identify those who may benefit from frailty-focused and depression-focused interventions. However, although the models I de- veloped have good predictive performance overall, they were found to have unequal performance by patient race and gender. Additionally, there is a large evidence gap around the feasibility and acceptability of using such machine learning models to guide clinical care in this setting. Therefore, the work proposed in this R03 will fill these important knowledge gaps and pave the way for a future randomized clinical trial of a predictive clinical decision support system to promote population health among people living with chronic lung disease. First, I will use state-of-the-art methods to recalibrate the prediction models so that they perform equitably by patient racial and gender subgroups. This aim is necessary, in light of known racial and disparities in care pro- cesses and outcomes by patient race and gender, to meet minimum ethical standards prior to further model deployment and testing. Second, I will conduct a pilot study of using these models in a real-world clinical workflow to prompt referrals among community-dwelling patients with chronic lung disease. A pulmonologist will review the predictions from these models, verify their clinical appropriateness, and make referrals if the patient and their outpatient clinical team is agreeable. I will assess the feasibility and acceptability of this workflow among patients, their caregivers, and outpatient clinicians. Additionally, I will measure several key process measures related to the proposed protocol in anticipation of a future clinical trial. The findings from these two aims will be sufficient to support a future, planned R01-level application for a clinical trial of this predictive clinical decision support system. Finally, securing this funding, with the data produced through the work outlined in this R03 proposal, will support my transition to an independent investigator.
